A role for the Saccharomyces cerevisiae ABCF protein New1 in translation termination/recycling

Abstract Translation is controlled by numerous accessory proteins and translation factors. In the yeast Saccharomyces cerevisiae, translation elongation requires an essential elongation factor, the ABCF ATPase eEF3. A closely related protein, New1, is encoded by a non-essential gene with cold sensitivity and ribosome assembly defect knock-out phenotypes. Since the exact molecular function of New1 is unknown, it is unclear if the ribosome assembly defect is direct, i.e. New1 is a bona fide assembly factor, or indirect, for instance due to a defect in protein synthesis. To investigate this, we employed yeast genetics, cryo-electron microscopy (cryo-EM) and ribosome profiling (Ribo-Seq) to interrogate the molecular function of New1. Overexpression of New1 rescues the inviability of a yeast strain lacking the otherwise strictly essential translation factor eEF3. The structure of the ATPase-deficient (EQ2) New1 mutant locked on the 80S ribosome reveals that New1 binds analogously to the ribosome as eEF3. Finally, Ribo-Seq analysis revealed that loss of New1 leads to ribosome queuing upstream of 3′-terminal lysine and arginine codons, including those genes encoding proteins of the cytoplasmic translational machinery. Our results suggest that New1 is a translation factor that fine-tunes the efficiency of translation termination or ribosome recycling.


INTRODUCTION
The ribosome is aided and regulated by accessory factors participating in all stages of the translational cycle: initiation, elongation, termination and recycling (1)(2)(3)(4). Translational GTPases are the key players in all these steps and as such have been extensively studied over the past five decades (5,6). Another class of NTPase enzymes has attracted increasing attention in the recent years--the ribosomeassociated adenosine triphosphatases (ATPases) belonging to the ATP-binding cassette (ABC) type F (ABCF) protein family. Eukaryotic representatives of this protein family include initiation factor ABCF1/ABC50 (7,8), elongation factor eEF3 (9,10) as well as Gcn20--a key component of general amino acid control pathway (11).
Elongation factor eEF3 is the most well-studied eukaryotic ribosome-associated ABCF ATPase. While early analyses of the evolutionary distribution of eEF3 concluded that it is a fungi-specific translational factor (12) eEF3like homologues have more recently been found in nonfungal species, such as oomycete Phytophthora infestans (13), choanoflagellates and various algae (14). eEF3 is essential both for viability of the yeast Saccharomyces cerevisiae (10) and for peptide elongation in a reconstituted yeast translational system (15,16). While biochemical experiments suggest a secondary function for eEF3 in ribosome recycling (17), ribosome profiling analysis of eEF3-depleted S. cerevisiae suggests that translation elongation is the primary function of eEF3 in the cell (18). A cryoelectron microscopy (cryo-EM) reconstruction of eEF3 on the ribosome localizes the factor within the vicinity of the ribosomal E-site (9), providing a structural explanation for the biochemical observations of E-site tRNA displacement from the ribosome in the presence of eEF3 and ATP (19).
The S. cerevisiae genome encodes 29 ABC ATPases, of which five belong to the ABCF subfamily (20). Of these, eEF3, Hef3/eEF3B and New1 together form a group with a distinctive subdomain architecture relative to the other ABCFs (14). Hef3/eEF3B is the closest homologue of eEF3--84% identical on the amino acid level (21)--and is, most likely, a second copy of eEF3 originating from wholegenome duplication in the Saccharomyces lineage (22). The second closest homologue is New1 (20)--encoded by a nonessential gene with a cold sensitivity knock-out phenotype (23) (Figure 1A). The molecular function of New1 is unclear. eEF3 and New1 share the same domain architecture, with exception of an additional N-terminal Q/N rich region in New1 that drives the formation of the [NU + ] prion when fused to the C-terminal domains of translation termination factor eRF3 (24,25) ( Figure 1A). However, prionogenesis cannot be the sole function of New1 since the very presence of the Q/N motif in New1 is not universal and is limited to Hemiascomycota species (26), and is lacking, for instance, in Schizosaccharomyces pombe New1 (14) ( Figure  1A). Loss of New1 results in a ribosome assembly defect in S. cerevisiae and causes cold sensitivity, that is, a growth defect at low temperatures (23). However, it is unclear whether the effect is direct, through participation of New1 in ribosome assembly, or indirect, as perturbation of translation can cause ribosome assembly defects (27). The latter possibility is supported by the detection of New1 in polysomal fractions (23), motivating our current investigation.
Using a combination of ribosome profiling, yeast genetics, molecular biology and cryo-EM, we show that S. cerevisiae New1 is a yeast translational factor involved in translation termination/recycling and its loss leads to ribosome queuing at C-terminal lysine and arginine residues that precede stop codons.

Yeast genetics
Yeast strains and plasmids used in this study are listed in Table 1. Oligonucleotides and synthetic DNA sequences are listed in Supplementary Table S1 and construction of plasmids is described in the Supplementary information. Yeast media was prepared as described (28), with the difference that the composition of the drop-out mix was as per Johansson (29). Difco Yeast Nitrogen Base w/o Amino Acids was purchased from Becton Dickinson (291 940), amino acids and other supplements from Sigma-Aldrich. YEPD medium supplemented with 200 g/ml Geneticin™ (Gibco 11811-023) was used to select for transformants containing the kanMX6 marker (30). Strains deleted for NEW1 were derived from a diploid formed between BY4741 and BY4742, which was transformed with either a HIS3MX6 or kanMX6 DNA fragment with the appropriate homologies to the NEW1 locus. The DNA fragments were ampli-fied from pFA6a-HIS3MX6 and pFA6a-kanMX6 (30) using primers oMJ489 and oMJ490. The two generated heterozygous strains were allowed to sporulate and the new1Δ strains (MJY944, MJY945 and MJY1091) were obtained from tetrads. The new1Δ::kanMX6 strain (MJY1091) was crossed to BY4709 and strains MJY1171 and MJY1173 were obtained from a tetrad. The P MET25 -YEF3 new1Δ strain (MJY951) was derived from a cross between VKY8 and MJY944. A strain expressing the synthetic LexA-ER-haB112 transcription factor (VHY61) was constructed by integrating the FRP880 plasmid (31) into the his3Δ1 locus of BY4741. To target the plasmid to the his3Δ1 locus, the plasmid was linearized with PacI before transformation.

Polysome profile analyses
Polysome profile analyses and fractionation was performed as described earlier (18). To probe the ribosomal association of New1, the VHY61 strain carrying the ␤estradiol-inducible NEW1-TAP (VHp262) or NEW1-EQ 2 -TAP (VHp265) plasmid was grown overnight at 30 • C in SC-ura and diluted to OD 600 ≈ 0.05 in the same medium. Cultures were grown until OD 600 ≈ 0.3 at which point ␤estradiol was added to a final concentration of 1 M. Cells were harvested after 3 h of induction and lysates prepared as previously described (18). The lysis buffer and sucrose gradient were supplemented where appropriate with ATP or AMP-PNP to a final concentration of 1 mM. RNase I treatment was performed for 20 min at room temperature before loading on a 7-45% sucrose gradient in of HEPES:Polymix buffer (20 mM HEPES:KOH pH 7.5, 2 mM dithiothreitol (DTT), 10 mM Mg(OAc) 2 , 95 mM KCl, 5 mM NH 4 Cl, 0.5 mM CaCl 2 , 8 mM putrescine, 1 mM spermidine (32)) and resolving the samples by centrifugation at 35 000 rpm for 3 h at 4 • C. Since 1 mM ATP masks the ribosomal absorbance signal at 260 nm, the gradients were analysed with continuous A 280 measurements. 0.5 ml fractions were collected and stored at −20 • C for further analysis.

Immunodetection
The TAP tag was detected using Peroxidase-anti-peroxidase 1:5000 (Sigma, Lot #103M4822), Pgk1 was detected using mouse anti-Pgk1 1:7500 (459250, Invitrogen). Anti-mouse IgG HRP (AS11 1772, Agrisera) was used as secondary antibody. For western blotting of fractions from polysome profiles, 200 l of Polysome Gradient Buffer without sucrose (20 mM Tris-HCl pH 7.5, 50 mM KCl, 10 mM MgCl 2 ) was added to 200 l of each fraction and the samples precipitated by addition of 1.2 ml of 99.5% EtOH and overnight incubation at −20 • C. The samples were pelleted by centrifugation at 20 000 × g for 30 min at 4 • C, air dried for 10 min at 50 • C and dissolved in 24 l of 3× Sample Loading Buffer (125 mM Tris-HCl pH6.8, 50% glycerol, 1.43M ␤-mercaptoethanol (BME), 4% sodium dodecyl sulphate (SDS), 0.025% Bromophenol Blue). A total of 5 l of each sample was resolved by 8% SDS-polyacrylamide gel electrophoresis. Since the detection of the TAP-tag is highly specific (see uncut western blots presented on Supplementary Figure S1), we increased throughput by slot blotting. Slot blotting of fractions from polysome profiles was performed using a Minifold II Slot-Blot Manifold (Schleicher  Figure 1. The essential translational factor eEF3 is dispensable upon overexpression of New1. (A) The domain structure of Saccharomyces cerevisiae and Schizosaccharomyces pombe New1 as well as S. cerevisiae eEF3. The location of the catalytic glutamate residues essential for ATPase function is indicated.
(B) Loss of NEW1 increases the growth defect caused by depletion of eEF3. The wild-type (VKY9), P MET25 -YEF3 (VKY8), new1 (MJY945) and P MET25 -YEF3 new1 (MJY951) strains were grown overnight in liquid SC-met-cys medium, 10-fold serially diluted, spotted on SC-met-cys (0 mM met) and SC-cys (0.5 mM met) plates. The plates were scored after three days incubation at 30 • C. (C) Increased dosage of NEW1 counteracts the growth defect caused by reduced eEF3 levels. The P MET25 -YEF3 strain carrying the indicated low-copy (l.c.) or high-copy (h.c.) URA3 plasmids was grown overnight in liquid SC-ura-met-cys medium, 10-fold serially diluted, spotted on SC-ura-met-cys (0 mM met) and SC-ura-cys (2 mM met) plates, and incubated at 30 • C for 3 days. (D) Increased expression of NEW1 counteracts the inviability of cells lacking eEF3. The yef3 strain harbouring the l.c. URA3 plasmid pRS316-YEF3 (VKY20) was transformed with the indicated l.c. or h.c. LEU2 plasmids. The transformants were grown over-night in liquid SC-leu medium, 10-fold serially diluted, spotted onto SC-leu and SC-leu+5-FOA plates, and incubated at 30 • C for three (SC-leu) or 5 (SC-leu+5-FOA) days. On 5-FOA containing plates only those cells that have lost the URA3 plasmid are able to grow (74). (E) Increased dosage of the YEF3 gene does not suppress the growth defect of new1Δ cells. The cells harbouring the indicated plasmids were grown overnight in liquid SC-ura medium, 10-fold serially diluted, spotted on SC-ura plates and incubated at 30 • C for 2 or at 20 • C for 4 days.
& Schuell) at 30 mbar vacuum. The slots were washed with 150 l of Polysome Gradient Buffer, loaded with 150 l of the same buffer followed by addition of 10 l of each fraction in separate slots. After two washes with 150 l of Polysome Gradient Buffer, the membrane was transferred to a hybridization bottle and blocked, using 5% skimmed milk in PBS-T, for 2 h in a hybridization oven set at room temperature. Following antibody incubation (TAP-tag detection), washing and chemiluminescent detection was performed as for western blot (18).

Purification of 80S ribosomes
The S. cerevisiae strain PY116 was used for preparation of ribosomes (33). The overnight culture was grown in a fermenter (SP Processum, Domsjö) in 25 L glycerol-lactate medium to OD 600 of ≈ 5, followed by the addition of 25 L of YPD. After 5 h the culture reached the OD 600 of ≈ 5, and it was cooled down to 4 • C for several hours, the cells were pelleted by centrifugation, flash-frozen and stored in −80 • C.  (32)). The 80S concentration was measured spectrophotometrically (1 A 260 = 20 nM of 80S). Ribosomes were aliquoted by 50-100 l, flashfrozen in liquid nitrogen and stored at −80 • C.

Cryo-electron microscopy and single-particle reconstruction
Data collection was performed on an FEI Titan Krios transmission electron microscope equipped with a Falcon II direct electron detector (FEI) at 300 kV at a pixel size of 1.063Å. A total of 3519 Micrographs were collected using an under defocus range from −0.5 to −2.5 m. Each micrograph was collected as series of 16 frames (2.7 e − /Å 2 pre-exposure; 2.7 e − /Å 2 dose per frame). All frames (accumulated dose of 45.9 e − /Å 2 ) were aligned using the Unblur software (34). Power-spectra, defocus values, astigmatism and resolution estimates for each micrograph were determined using Gctf version 1.06 (35). After manual inspection and application of a resolution cut-off which preserved micrographs showing Thon rings beyond 3.5Å, 1919 micrographs were subjected to the next processing steps. Automated particle picking was performed using Gautomatch version 0.56 (http://www.mrc-lmb.cam.ac.uk/kzhang/) resulting in 171 106 particles, which were first processed using RELION-2.1 (36). Manual inspection of 6xbinned 2D particle classes after 2D classification resulted in 144 876 particles that were further used for 3D-refinement using an empty S. cerevisiae 80S ribosome (density map was created using the PDB ID: 3j78 with the tRNAs removed) as a reference structure. Subsequently, a 3D classification was performed sorting the particles into five classes (Supplementary Figure S2). The class containing the New1-80S complex (57 226 particles) was subjected to focusedsorting, using a mask encompassing the New1 ligand. The New1-containing population (48 757 particles) was than refined using undecimated particles and additionally focused refined with a spherical mask encompassing the New1 ligand as well as the 40S head (Supplementary Figure  S2). Subsequently, the overall refined structure was subjected to contrast-transfer function (CTF) refinement using RELION-3.0 (37). The CTF refined particles were again 3D refined and subsequently multibody refinement was performed. For the multibody refinement, three masks were used: the first one encompassed the first portion of New1 (HEAT and 4HB, residue range 141-549) as well as the SSU Head, the second mask included the remain part of New1 (ABC1-ABC2-CD, residue range 570-1112) and the LSU, and the third mask covered the SSU Body (Supplementary Figures S2 and 3E). The final reconstructions were corrected for the modulation transfer function of the Falcon 2 detector and sharpened by applying a negative B-factor estimated by RELION-3.0 (37). For the sharpening, a mask for the whole New1-80S complex (for the overall refinement) and New1 (for the local refinement) was used, respectively, resulting in a final average reconstruction of 3.28Å for the New1-80S complex and 3.63Å for the New1 (Supplementary Figures S2 and 3A). The same was done for each part of the multibody refined volumes, which provided an average resolution of 3.26Å for the SSU Head-New1 (HEAT, 4HB), 3.06Å for the LSU-New1 (ABC1, ABC2, CD) and 3.14Å for the SSU Body. The single multibodies were merged using phenix 1.15 via the phenix.combine focused maps command (38). The resolution of the 3.28Å volume was es-timated using the 'gold standard' criterion (FSC = 0.143) (39). Local-resolution estimation and local filtering of the final volumes was done using Relion-2.1 (Supplementary Figure S3F-N).

Molecular modelling
A homology model of New1 was created using the deposited structure of the eEF3 protein model (PDB ID: 2IX8) (9) as a template for SWISS-MODELLER (40). The homology model was fitted into density with UCSF Chimera 1.12.1 (41) using the command 'fit in map': Single domains were manually adjusted with Coot version 0.8.9.2 (42). The chromo domain was built de novo based on the secondary structure elements of the chromo domain of HP1 complexed with the histone H3 tail of Drosophila melanogaster (PDB ID: 1KNE). The ATP molecules within the ABCs were obtained from the PDB ID: 6HA8 (43). The model of the S. cerevisiae 80S ribosome was derived from PDB ID: 4U4R. The proteins of the SSU and LSU were fitted separately into locally filtered electron density maps using UCSF Chimera (41).
The rRNA was fitted domain-wise in Coot (42). Afterwards manual adjustments were applied to all fitted molecular models using Coot. The combined molecular model (proteins+rRNA) was then refined into the merged multibody maps using the phenix.real space refine command of phenix version 1.14, with restraints that were obtained via the phenix.secondary structure restraints command (38). Cross-validation against overfitting was performed as described elsewhere (44) (Supplementary Figure S3B). Statistics for the model were obtained using MolProbity (45) and are represented in Supplementary Table S2.

Figure preparation
Figures showing atomic models and electron densities were generated using either UCSF Chimera (41) or Chimera X (46) and assembled with Inkscape.

Preparation of Ribo-Seq and RNA-Seq libraries and data analysis
Cultures of wild type (MJY1171) and new1Δ (MJY1173) strains were grown overnight at 20 • C (or, alternatively, at 30 • C), diluted to OD 600 ≈ 0.05 in 750 ml of SC medium and incubated in a water bath shaker at 20 • C (or 30 • C) until OD 600 ≈ 0.6. Cells were harvested by rapid vacuum filtration onto a 0.45 m nitrocellulose membrane, scraped off, and frozen in liquid nitrogen. Cells were lysed by cryogenic milling and RNA-Seq and Ribo-Seq libraries prepared from the cell extracts. RNA-Seq libraries were prepared using Scriptseq Complete Gold Yeast Kit (Epicentre). Ribo-Seq libraries were prepared essentially as per Ingolia and colleagues (18) with updated enzymes, modifications of rRNA removal and sample purification procedures. Detailed protocols can be found in our GitHub repository https://github.com/GCA-VH-lab/Lab protocols.
In the case of Ribo-Seq, out of 36-47 million unprocessed reads after removal of non-coding RNA (rRNA and tRNA) and reads that mapped more than twice 3-8 million reads remained. Pooling resulted in 7.8 (20 • C)/9.3 (30 • C) million reads total for wild-type and 12.7 (20 • C)/5.5 (30 • C) million reads for new1Δ. RNA-Seq reads were processed similarly, omitting the Cutadapt step; out of 16-32 million unprocessed reads, 12.5-20 million remained after removing non-coding RNA and reads that mapped more than two times. The analysis pipeline was implemented in Python 3 and is available at GitHub repository (https://github.com/GCA-VH-lab/2019-NAR). Reads that mapped once to the genome were used for final P-site assignment using read length-specific offsets, as provided in the file readlength offsets expert.txt on GitHub. Mapped reads were normalised as reads per million (RPM). The ribosome queuing metric was computed using the Python script compute queuing.py. Gene coverage was calculated using HTSeq-count (50) and differential gene expression (DE) analyses were performed using the DESeq2 package (51). All four replicates (two at 30 • C and two at 20 • C) of wild type and new1Δ datasets were used in the analyses. Differentially expressed genes were identified based on an adjusted P-value cut-off of 0.05 as well as at least 2-fold difference in expression level (Supplementary Table S4). Only the genes that with >10 Ribo-Seq or RNA-Seq were used for analysis. GO enrichment analysis was performed using the Gene Ontology (GO) knowledgebase (http://geneontology. org/) (52) implemented in PANTHER (53). The R script used for analyses, as well as both input and output files are available at GitHub repository (https://github.com/GCA-VH-lab/2019-NAR/tree/master/Differential Expression).

New1 has a function that partially overlaps with that of eEF3
New1 s closest homologue in S. cerevisiae--the translation factor eEF3 encoded by the YEF3 gene--is an essential player in translation elongation (15,16) with a putative secondary function in ribosome recycling (17). While deletion of the NEW1 gene alone causes a pronounced growth defect (40% reduction) at 20 • C (23), the growth defect at 30 • C is relatively minor (10%) (Supplementary Table S3). To investigate a potential overlap in the functions of New1 and eEF3, we introduced a new1Δ allele into cells in which YEF3 is under the control of the methionine repressible MET25 promoter (P MET25 ) (54). We have previously characterized the P MET25 -YEF3 strain and demonstrated that addition of increasing methionine concentrations specifically leads to a decrease in the eEF3 levels, resulting in a gradual slow-down of growth (18) (Supplementary Figure S4A). The methionine-induced growth inhibition of P MET25 -YEF3 cells was more pronounced in the absence of NEW1, at both 20 • C and 30 • C ( Figure 1B and Supplementary Figure S4BC). While these results are consistent with overlapping functions of New1 and eEF3 in protein synthesis, the additive negative effects on fitness could also be due to the two proteins acting on two sequential steps in gene expression, for example, ribosome assembly (New1) and translation (eEF3). Therefore, we investigated whether New1 can functionally substitute for eEF3 s cellular function. To do this, we first tested whether full-length New1 can supress the growth defect caused by eEF3 depletion. We found that increased NEW1 gene dosage did, indeed, counteract the growth retardation caused by eEF3 depletion in P MET25 -YEF3 cells ( Figure 1C and Supplementary Figure  S4D). Second, we investigated whether an increased level of New1 could overcome the essentiality of eEF3. To do so, we constructed a yef3Δ strain that was rescued by a wild-type YEF3 gene on a URA3 plasmid. Increased dosage of the NEW1 gene, either under its own or the strong TDH3 promoter (P TDH3 ), counteracted the requirement of the YEF3 plasmid for viability ( Figure 1D). Taken together, our findings demonstrate that New1 can, although with low efficiency, perform the essential molecular function of eEF3 in protein synthesis. On the other hand, increased YEF3 gene dosage does not suppress the cold sensitivity phenotype of a new1 strain, suggesting that New1 has a specific molecular function that cannot be performed by eEF3 ( Figure 1E).

The ATPase-deficient New1-EQ 2 mutant is locked on translating ribosomes
To characterise the association of New1 with ribosomes, we used ATPase-deficient mutants in which the catalytic glutamic acid residues (E713 and E1058) in the two ATPase active sites ( Figure 1A) are replaced by glutamines (Q) (socalled EQ 2 mutants). Locked in the ATP-bound conformation, the EQ 2 mutants of E. coli ABCF EttA (55), Bacillus subtilis VmlR (43) and human ABCF ABCF1/ABC50 (7) were previously used to trap these ABCF ATPases on ribosomes. We used a tightly controlled ␤-estradiol-inducible LexA-ER-B112 expression system (31) to drive expression of C-terminally TAP-tagged (56) versions of New1 or New1-EQ 2 . While induction of New1-TAP has no effect on growth, the induction of New1-EQ 2 -TAP leads to growth inhibition ∼3 h after the addition of the inducer ␤-estradiol (Figure 2A and Supplementary Figure S5AB). At this time point, the level of New1-EQ 2 is about 3-fold higher than those observed in a strain that expresses a Cterminally TAP-tagged New1 from its normal chromosomal location (Supplementary Figure S1). One likely reason for the growth inhibition is the non-productive association of New1-EQ 2 with its ribosomal target, be it ribosome assembly intermediates or mature 80S ribosomes. We directly tested this hypothesis by performing polysome fractionation and immunoblotting. In ATP-supplemented lysates, the bulk of New1-EQ 2 --but not wild type New1--localizes to the polysomal fractions, suggesting that translating ribosomes are indeed the cellular target of the factor (Figure 2B and Supplementary Figure S5C). Omitting ATP from the lysis buffer and sucrose gradients leads to loss of New1-EQ 2 from polysomal fractions, suggesting that ATP  Figure S5D). The interaction with ribosomes is specific, since collapsing polysomes either by RNase I treatment or preparing yeast lysates under run-off conditions (that is, in the absence of cycloheximide), leads to loss of New1-EQ 2 from the high-sucrose fractions and redistribution into the 80S fraction (Supplementary Figure S5F-G). Unlike native ATP, a commonly used non-hydrolysable analogue AMP-PNP fails to stabilize wild-type New1 binding to polysomes (Supplementary Figure S5H), suggesting that AMP-PNP is a poor substrate for New1 ( Figure 2B). Therefore, for solving the structure of New1 locked on the 80S ribosome, we used the EQ 2 mutant and ATP instead of using the wild-type protein and AMP-PNP.

Structure of the New1 Q/N -EQ 2 on the 80S ribosome
To determine the binding site of New1 on the ribosome, we incubated S. cerevisiae 80S ribosomes with New1 Q/N -EQ 2 variant in the presence of ATP, and subjected the assembled New1 -Q/N -EQ 2 -80S complex (hereafter referred to as the New1-80S complex) to single particle cryo-EM ( Figure  3A). After 2D classification, a total of 144 876 ribosomal particles were sorted into five distinct ribosomal subpopulations (Supplementary Figure S2). One major population (class 1 containing 57 226 particles, 39.5%) was obtained that contained additional density spanning the head of the 40S subunit and central protuberance of the large 50S subunit, whereas this density was either fragmented (class 2) or absent (class 3-5) in the remaining subpopulations (Supplementary Figure S2). Further local 3D classification of class 1 and the subsequent refinement of the defined New1-80S particles (class 2 containing 48 757 particles) yielded a final reconstruction of the New1-80S complex ( Figure 3A), with an average resolution of 3.28Å (Supplementary Figures S2  and 3). Local resolution calculations reveal that while the core of the ribosomal subunits was well resolved and ex-tended to 3.0Å, the peripherally bound New1 ligand was less well-resolved, with local resolution ranging between 4-7Å (Supplementary Figure S3C-F). To improve the overall resolution of the New1 density, further focussed and multibody refinement were performed (see 'Materials and Methods' section). The resulting maps showed significant improvements in the local resolution for New1, particularly at the interaction interface between New1 and 80S ribosome where the resolution approached 3Å ( Supplementary Figure S3G-N) and sidechains were clearly visible, whereas the periphery was less well-resolved (4-7Å) and only large and bulky sidechains could be visualized ( Supplementary Figure S6A and B). Together with the high sequence homology to eEF3, this enabled us to generate homology models for each domain of New1 using the crystal structure of eEF3 (9) as a template, which could then be unambiguously fitted and refined into the cryo-EM map of the New1-80S complex ( Figure 3B and C; Supplementary Figure S6C). Extra density was also observed for two molecules of ATP bound within each of the active sites of the ABC1 and ABC2 domains of New1 (Supplementary Figure S6D and E). This is consistent with the ability of New1-EQ 2 variant to bind, but not hydrolyse, ATP. The ABC1 and ABC2 nucleotide binding domains (NBDs) were observed in a closed conformation, analogous to the closed conformations adopted in the 70S-bound ABCF protein VmlR from B. subtilis (43) and archaeal 40S-ABCE1 complex (57)

Interaction of New1 with the 80S ribosome
To ascertain the contacts between New1 and the 80S ribosome, the crystal structure of the S. cerevisiae SSU and LSU was also fitted and refined together with New1 (Figure 3B). Overall, the binding site is globally similar to that observed previously for eEF3; however, with the improved resolution it is now possible to better ascertain which regions of New1 interact with which ribosomal components ( Figure 3D-F). The majority of the contacts with the SSU are established between the N-terminal HEAT domain and components of the SSU head, specifically, with the tip of expansion segment 9 (ES9) of the 18S rRNA and ribosomal proteins uS13 and uS19 ( Figure 3D-F). The 4HB and ABC1 domains of New1 do not interact with any ribosomal components, whereas ABC2 and CD form extensive interaction with the LSU, mainly with the 5S rRNA and ribosomal protein uL5. Additionally, bridging interactions are formed by ABC2 that contacts uL5 on the LSU and uS13 on the SSU ( Figure 3D-F). A more in-depth description of these interactions is found in the Supplementary Results section, with an accompanying Supplementary Figure S8. By contrast, the tip of the CD makes no contacts with the ribosome, but rather extends into the vacant E-site in the direction of the L1 stalk ( Figure 4A), as observed previously for the eEF3-80S structure ( Figure 4B) (9). The CD of eEF3 has been proposed to influence the conformation of the L1 stalk and thereby facilitate release of the E-site tRNA (9). However, compared to eEF3, the CD of New1 is truncated and lacks the ␤4-␤5-hairpin ( Figure 4C-D) and therefore the extent by which the CD of New1 reaches towards the L1 stalk is reduced ( Figure 4A). We speculate that the ␤4-␤5-hairpin plays an important role in eEF3 function to promote the open conformation of the L1 stalk necessary for E-tRNA release and that the absence of this motif in New1 may explain why New1 overexpression cannot fully rescue the growth defect caused by depletion or loss of eEF3. This raises the question as to whether New1 has acquired another function on the ribosome that is distinct from eEF3.

New1 depletion leads to ribosome queuing upstream of 3terminal lysine and arginine codons
To assess the global effects of New1 loss on translation in yeast we used ribosome profiling (Ribo-Seq) (59) of wildtype and new1Δ strains both at 20 • C and 30 • C, with two biological replicates for each condition. We also carried out RNA-seq to enable normalization of translation level for the transcription level effects on expression. We performed differential expression analysis of both RNA-Seq (38 genes with increased coverage and 48 genes with decreased coverage) and Ribo-Seq (26 genes with increased  Table S4, see 'Materials and Methods' section for details). Next, we applied Gene Ontology (GO) analysis (52) to identify the biological processes, molecular functions and cellular components enriched in the differentially expressed genes (Supplementary Table S4). Increased RNA-Seq coverage is associated with mitochondrial function and cell energetics, such as ubiquinol to cytochrome c (GO:0006122), ATP synthesis coupled proton transport (GO:0015986), ATP hydrolysis coupled cation transmembrane transport (GO:0099132) and mitochondrion organization (GO:0007005); analysis of the genes with differentially increased Ribo-Seq coverage reveals no GO enrichment. Decreased RNA-Seq coverage is associated with urea cycle (GO:0000050), ethanol biosynthetic process involved in glucose fermentation to ethanol (GO:0043458), arginine biosynthetic process (GO:0006526), NADH oxidation (GO:0006116) and amino acid catabolic process to alcohol via Ehrlich pathway (GO:0000947). The genes with decreased Ribo-Seq coverage are associated with an overlapping set of GO terms, suggesting the effects are predominantly at the mRNA level. The Ribo-Seq procedure generates two distinct classes of RPFs: 'short' (20-22 nucleotides) and 'long' (28-30 nucleotides), which originate from ribosomes with vacant or occupied A-sites, respectively (60). The translation elongation inhibitor cycloheximide has previously been a common component in Ribo-Seq library preparation; however, this antibiotic has recently been found to drive near-excusive formation of long RPFs (60,61). Therefore, to be able to use the RPF size distribution as an additional insight into the nature of translational perturbation upon the loss of New1, we omitted cycloheximide during cell collection and lysis.
At 30 • C the RPF length distributions of wild-type and new1Δ are very similar (Supplementary Figure S9A). However, at 20 • C, i.e. when New1 absence leads to a significant growth defect (Supplementary Table S2), short RPFs are moderately depleted in the new1Δ dataset (Supplementary Figure S9B). While depletion of short RPFs signals a lower abundance of ribosomes with a vacant A-site, the causality is unclear: the lower fraction of the 80S with an empty Asite could reflect a perturbation of translation in the absence of New1 or be a mere consequence of the slower growth rate. Strikingly, the lack of New1 leads to the appearance of periodic (≈30 nt) 'waves' of ribosomal density preceding the stop codon ( Figure 5A and Supplementary Figure  S9C-F). The period of the 'waves' corresponds to the size of mRNA fragment protected by a ribosome, suggesting ribosome queuing upstream of the stop codon due to a defect in translation termination or/and ribosome recycling. The effect is more pronounced at 20 • C, correlating with the severity of the growth defect (compare Supplementary Figure S9C-F). We compared our Ribo-Seq analysis of the new1Δ strain with other yeast strains defective in translation termination, namely, S. cerevisiae depleted for eIF5A (62)  new1Δ strain is much more dramatic and no queuing is detectable in the [PSI+] Ribo-Seq dataset. Our metagene plots lack the pronounced peak at the stop codon, both in the case of wild type and new1Δ datasets. This is not unexpected as ribosomal post-termination complexes are more sensitive to ionic strength than elongating ribosomes (62) and require stabilisation by cycloheximide that was specifically omitted in our Ribo-Seq protocol (65). We next set out to determine whether only a subset of translated mRNAs is affected by New1 loss--and if this is the case, which ORF features correlate with aberrant mRNA translation in the new1Δ strain. We defined the 'Cterminal ribosome queuing metric' (or just 'queuing metric' for short) as a ratio between the sum of the maximal Ribo-Seq counts within the three peaks preceding the stop codon (highlighted with grey shading on Figure 5A) and the average Ribo-Seq density between the peaks. We plotted distributions of the queuing metric values for individual datasets ( Figure 5C) as well as for individual ORFs in new1Δ against wild-type (Supplementary Figures S11A). While the effect of New1 loss on the mean queuing metric is moderate (5.6 wt versus 9.6 new1Δ), the distribution in the new1Δ dataset has a clear 'heavy tail'. While the high queuing metric in new1Δ is associated with relatively higher ribosome density after the stop codon (3 UTR) (Supplementary Figure   S12A-F), the 3 UTR density in genes displays no codon periodicity suggesting that new1Δ does not lead to increased readthrough (Supplementary Figure S12C and F). Consistent with this notion, analyses of readthrough using a dualluciferase readthrough reporter system (66) detected no apparent difference between the new1Δ and wild-type strains (Supplementary Figure S17C).
We selected the ORFs with high queuing metric values using a Z-score cut-off of 1 and applied pLogo (67) to analyse the sequence conservation pattern in the C-terminal region encompassing five codons preceding the stop codon. Strikingly, in the case of new1Δ the selected ORFs display a clear conservation signal localized at the C-terminal amino acid, dominated by positively charged lysine (K) and arginine (R) residues ( Figure 5D). In the wild-type dataset there is no signal passing the significance cut-off; and no specific C-terminal amino acid signal is detectable in eIF5A (62) or Rli1/ABCE (63) datasets (Supplementary Figure  S13), which could be attributed to the relative weakness of the queuing signal in these datasets. Therefore, we tested if the nature of the C-terminal amino acid corelates with the increase in the ribosomal density at the stop codon upon eIF5A and Rli1/ABCE depletion and in [PSI+] (Supplementary Figure S14). Our analysis detects no specific increase of the Ribo-Seq density at the stop codons preceded by C-terminal arginine and lysine residues, further reinforcing the specificity of the effect we observed in the new1Δ strain. Finally, we tested if the nature of the stop codon is associated with ribosomal queuing, but detected no signal (Supplementary Figure S15). At the same time, stop codon context analysis readily picks up an over-representation of C-terminal codons encoding lysine (AAA) and arginine (AGA and CGN), consistent with an effect on the amino acid level.
After establishing that the nature of the C-terminal amino acid is the key predictor of ribosomal queuing at the C-terminus, we plotted the mean queuing metric for ORFs grouped by the nature of the different C-terminal amino acid ( Figure 5E; note the difference in scale of the X and Y axis). Importantly, in this case we have not pre-selected a subset of ORFs but rather analysed all of the 876 ORFs with sufficient coverage of 3 -terminal 120-nucleotide region (≥10 rpm) in all the four conditions (wt and new1Δ, 20 • C and 30 • C). The analysis readily picks up lysine and arginine as outliers in the new1Δ background, and detects a weaker signal for asparagine (N). We plotted queuing metric distributions for individual ORFs grouped by the nature of the C-terminal amino acid (Lysine: Figure 5F; distributions for ORFs ending with other amino acids are presented on Supplementary Figure S16). In the case of wild-type, distribution of the queuing metric for ORFs terminating with lysine is similar to that of ORFs in general (compare Figure 5C and F). Conversely, new1Δ ORFs terminating with lysine display a broad distribution that is strongly shifted to higher values, consistent with significantly higher ribosomal queuing on mRNAs encoding C-terminal lysine residues ( Figure 5E).
After establishing that effect of the New1 loss is clearly mRNA-specific and is not solely determined by the Cterminal residue alone, we re-assessed the stop codon readthrough in new1Δ using a modified dual-luciferase reporter system. We chose two ORFs--YDR099W / BMH2 (C-terminal lysine, UAA stop codon) and YNL247 (Cterminal lysine, UAG stop codon)--which in new1Δ strain are characterized by possessing high queuing score and increased RiboSeq density in 3 UTR. Additionally, these mRNAs show sufficient RiboSeq coverage as per visual inspection (Supplementary Figure S17A and B). We constructed readthrough reporters containing the four codons immediately preceding and the two codons after the native stop codon present in YDR099W and YNL247. Despite containing C-terminal lysine residues, neither of the two reporters display an increased readthrough level in the new1 background (Supplementary Figure S17C).

DISCUSSION
Our results show a novel role for New1 in fine-tuning the efficiency of translation termination and/or ribosome recycling in S. cerevisiae. Two molecular functions have previously been documented for New1. The first is its necessity for efficient ribosome assembly, which is phenotypically manifested in a pronounced cold-sensitivity of the new1Δ strain (23). While New1 loss compromises ribosomal assembly, this phenotype does not necessitate that the direct role of New1 in the process is as a bona fide assem-bly factor. A homologue of New1 and the bacterial ABCF translation factor EttA (55)--Uup--presents an analogous case: while overexpression of Uup supresses both cold sensitivity and ribosome assembly in E. coli caused by knockout of the translational GTPase BipA (14), it is unclear if Uup is an assembly factor or translation factor. The association of New1 with polysomal fractions (see Figure  2B and (23)) indicate that the effect on ribosome assembly is indirect and is caused by the perturbation of translation in the new1Δ strain. Our yeast genetics, cryo-EM and Ribo-Seq results collectively argue that New1, indeed, acts on mature ribosomes, and its loss affects translation termination/recycling, leading to ribosomal pile up in front of stop codons preceded by a lysine or arginine codon. The genes involved in protein biosynthesis display only a mild enrichment in C-terminal arginines and lysines compared to ORFs in general (23 versus 21%, Supplementary Figure S11B). However, the metagene of translation-associated genes displays a clear ribosome stalling pattern (Supplementary Figure S11C). We conclude that the ribosome assembly defect in the new1Δ strain could be a knock-on effect of the protein synthesis defect, although it is also possible that New1 is a dual-function factor that is involved both in protein synthesis and ribosome assembly, similarly to Rli1/ABCE1 (68,69), eIF5B/Fun12 (70,71) and eIF6/TIF6 (72). A dedicated follow-up study is necessary to test this hypothesis. The second proposed molecular function of New1 is mediated by its prion-like propensity to aggregate in S. cerevisiae (24). When overexpressed, New1 influences formation of the prion state of translation termination factor eEF3/Sup35--[PSI+]--and the effect is strictly dependent on New1 s N-terminal Q/N rich region (25,73). Importantly, since the Q/N region is not universal for New1--the distribution of this region is limited to Hemiascomycota species (26)--prionogenesis is not a general feature of New1.
Here we demonstrate that New1 binds to ribosomes and plays a role in facilitating translation termination and/or ribosome recycling at stop codons preceded by a codon for the positively charged amino acids lysine or arginine. However, the mechanism by which New1 facilitates this process remains to be elucidated. At present, we can only speculate that binding of New1 to the ribosome in a position analogous to eEF3, exert its influence via the E-site, possibly by affecting the stability of the E-site tRNA and/or the conformation of the L1 stalk, however we also cannot rule out that New1 is involved in recruitment or dissociation of other auxiliary factors from the ribosome that directly influence the termination/recycling process. A systematic follow-up investigation is necessary to establish the connection between the termination / recycling defect caused by New1 loss.

DATA AVAILABILITY
Ribo-Seq and RNA-Seq sequencing data have been deposited in the ArrayExpress database at EMBL-EBI (www. ebi.ac.uk/arrayexpress) under accession number E-MTAB-7763. The cryo-EM maps of the New1-80S complex (including all individual multibody maps) and the associated molecular model have been deposited in the Protein Data Bank and Electron Microscopy Data Bank with the accession codes 6S47 and EMD-10098, respectively.